Quantum
Machine-learned active spaces enable quantum calculations of spin-state energetics of transition-metal complexes on quantum hardware
Publication: quant-ph updates on arXiv.org
Author: Muhammad Arsalan Ali Akbar, Thomas D. K\"uhne, Sabre Kais, Werner Dobrautz
Source published: 17/11/2026, 05:00:00 UTC
Briefing created: 17/11/2026, 08:40:55 UTC
Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 17/11/2026, 08:28:25 UTC.
Briefings
A new machine-learning model named SpinAware predicts DMRG single-orbital entropies and orbital-pair mutual information from unrestricted Hartree-Fock solutions, enabling quantum hardware to calculate spin-state energetics for transition-metal complexes. Trained on 3454 mononuclear Cr-Ni calculations, the model transfers to new complexes with different ligands, binding topologies, and oxidation states. It reproduces DMRG entropy profiles with a mean absolute error of 0.027 and identifies a candidate pool containing 99% of active orbitals while reducing orbital count to an average of 15. Sample-based quantum diagonalization on IBM processors matches CASCI energies within 0.3 kcal/mol and spin-state gaps within 0.10 kcal/mol for up to 32 qubits.
Key points
- The SpinAware model predicts DMRG single-orbital entropies and orbital-pair mutual information directly from unrestricted Hartree-Fock solutions without requiring any correlated calculations.
- The model was trained on 3454 spin-pure DMRG calculations for mononuclear Cr-Ni complexes and successfully transfers to experimentally characterized complexes with different ligands, binding topologies, and oxidation states.
Context
Machine-learned active spaces are crucial for quantum calculations of transition-metal complexes, as they must predict entanglement without performing the expensive correlated calculations they aim to avoid. This breakthrough enables quantum hardware to efficiently compute spin-state energetics for complex chemical systems.
Limitations
The model's performance is limited to mononuclear Cr-Ni complexes during training. While it transfers to new complexes, the specific ligands, binding topologies, and oxidation states tested during transfer were not part of the original training set. Quantum diagonalization results are based on samples from IBM processors and may vary with hardware specifics. The mean absolute error of 0.027 and gap errors of 0.10 kcal/mol are averages over 11 spin states and up to 32 qubits respectively.
Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 17/11/2026, 08:40:55 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.
View archived briefing on IPFS
bafybeid4zfn7u65kaq4lwhormojl7zmtzicsd7hogewcp5jh5wlpeuhady